The long-awaited checks have finally arrived for thousands of authors whose works were ingested into Anthropic's large language models, but the windfall has triggered an unexpected secondary conflict. Instead of a clean resolution to a copyright dispute, the payout process has devolved into a bitter fight between creators and the very publishers who once represented them. What began as a landmark legal battle over the ethics of AI training has shifted into a chaotic scramble over accounting records and the precise timing of contract terminations.

The Mechanics of a Billion Dollar Payout

Anthropic has entered into a massive $1.5 billion settlement to resolve class-action lawsuits brought by authors alleging their copyrighted works were used without permission. The scale of the agreement is vast, covering approximately 500,000 books. Under the terms of the settlement, the compensation is set at $3,000 per book, providing a tangible financial remedy to authors whose intellectual property fueled the development of generative AI. This settlement follows a pivotal judicial determination that attempted to draw a line between the act of learning and the act of copying. The presiding judge ruled that while using copyrighted material to train an AI model may fall under the principle of fair use—allowing for limited use without permission for research or educational purposes—the actual process of illegally reproducing those works to facilitate that training is a clear violation of the law.

This legal distinction provided the foundation for the settlement, which received final approval in July. The distribution of the funds is governed by the current status of the books' rights. For works that remain in active publication through traditional publishing houses, the settlement dictates a 50-50 split between the author and the publisher. However, the agreement provides a full 100 percent payout to the author in cases where the work was self-published or where the publisher has already returned the rights to the creator, effectively taking the book out of print.

The August 10 Deadline and the Record-Keeping Crisis

The transition from a legal victory to a financial payout has exposed a systemic failure in how the publishing industry manages its archives. The central point of contention is a single date: August 10, 2022. This date serves as the official download deadline, marking the point at which the data was collected for training. According to the settlement, whether an author is entitled to the full 100 percent of the payment depends entirely on whether the rights reversion process was completed before this specific cutoff. If the rights had already returned to the author by August 10, 2022, the publisher has no claim to the money. If the reversion happened after that date, the publisher is entitled to their half.

This rigid timeline has created a vacuum of uncertainty. Authors are now reporting instances where publishers are claiming a 50 percent share—or even the full 100 percent—of payments for books whose rights had long since been returned. Victoria Strauss has noted two distinct patterns of grievance: cases where publishers are claiming funds for works they no longer have any legal right to, and cases where they are demanding a higher percentage than the agreed-upon split. The conflict has extended beyond publishers to literary agencies. Some agencies, which typically act as intermediaries for contracts and sales rather than rights holders, have attempted to claim a portion of the settlement. Author Courtney Milan has been vocal in criticizing this behavior, arguing that agencies lack the legal standing to intercept funds intended for the copyright holders.

Mary Rasenberger, head of the Authors Guild, suggests that this chaos is not necessarily the result of a coordinated effort by publishers to steal from authors. Instead, she views the situation as a predictable consequence of abysmal record-keeping and an overly complex settlement structure. In her view, the industry's failure to maintain precise, digitized logs of rights reversions has collided with a high-stakes financial payout, turning administrative negligence into a legal crisis. The tension is no longer about whether AI companies should pay for data, but whether the legacy infrastructure of the publishing world is capable of accurately tracking who owns what in the age of algorithmic training.

The battle over AI training data has officially moved from the courtroom to the accounting office.